Fig 1.
Schematic representation of the StabilityCCA procedure.
Above are shown regularisation paths: the SCCA model as a function of sparsity. Below, stability paths, derived from the regularisation paths for 100 subsamples of size n/2. The stability score of a variable is the area-under-curve (AUC) of its stability path.
Fig 2.
Results for simulated data sets.
The base procedures are plotted with solid lines, while the dashed lines represent their different StabilityCCA extensions.
Fig 3.
The whiskers display one standard deviation for the test set correlations.
Fig 4.
Top-k model performance for different k values in the IBD data set.
The constant lines correspond to the average performance when an optimal value k* was selected through hyperparameter tuning.
Fig 5.
StabilityCCA models for species-metabolites IBD data.
(A) Top-50 model canonical variables. The canonical correlation (CC) is shown above the plot. (B) Top-10 model canonical variables. (C) Canonical coefficients and pairwise contributions to the canonical correlation for the top-10 model. (*) match to a standard with isomeric forms that could not be differentiated.
Table 1.
Top-10 variables from the species-metabolites data set.
Fig 6.
StabilityCCA models for enzymes-metabolites IBD data.
(A) Top-100 model canonical variables. The canonical correlation (CC) is shown above the plot. (A) Top-10 model canonical variables. (C) Canonical coefficients and pairwise contributions to the canonical correlation for the top-10 model. (*) match to a standard with isomeric forms that could not be differentiated.
Table 2.
Top-10 variables from the enzymes-metabolites data set.
Fig 7.
Canonical variables (A) and coefficients (B) for the species and enzymes views, taken from the top-10 species-metabolites model and the top-10 enzymes-metabolites model.